Thread by @xpvit

⚠️#Java performance tip⚠️

Don't use "+" for string concatenation. Use StringBuilder or StringBuffer instead.
I have created a benchmark to compare the speed of StringBuilder, StringBuffer and "+" concatenation for 2, 3, 5, 10, 50, and 100 strings.
Details in 🧵👇

The scenario is simple:
s/sb = ""/new StringBuffer/Builder();
for (int i = 0; i < size; i++) {
s+=i;/sb.append(i);
}
return sb.toString();

size = 2, 3, 5, 10, 50, and 100

Size hundred produced super strange result, which I can't explain yet. 😕

Concat 2 strings:
❎ s+= i; 25.528 ns/op ~15% slower
✅ stringBuilder.append(i); 21.265 ns/op
✅ stringBuffer.append(i); 21.251 ns/op

The difference is slight, so it is up to you to decide what to use. However, if you want the lowest latency, go with StringBuilder.

Concat 3 strings:
❌ s+= i; 42.559 ns/op ~40% slower
✅ stringBuilder.append(i); 25.461 ns/op
✅ stringBuffer.append(i); 25.810 ns/op

StringBuffer and StringBuilder are 40% faster then "+". 😀

Concat 5 strings:
❌ s+= i; 75.137 ns/op ~2x slower
✅ stringBuilder.append(i); 34.940 ns/op
✅ stringBuffer.append(i); 35.930 ns/op

WoW, StringBuffer and Builer concatenate 2x faster. And we concatenated only 5 strings. 😮

Concat 10 strings:
❌ s+= i; 157.276 ns/op 3x slower
✅ stringBuilder.append(i); 64.173 ns/op ~ 8% slower
✅ stringBuffer.append(i); 65.941 ns/op

10 strings concatenated. StringBuffer and StringBuilder Are 3x better than `s+=i;` 😎

Concat 50 strings:
❌ s+= i; 812.848 3x slower
✅ stringBuilder.append(i); 283.244 ns/op
✅ stringBuffer.append(i); 308.129 ns/op

It seems like we found pattern.

Concat 100 strings:
❌s+= i; 2090 ns/op 3.5x slower
✅ stringBuilder.append(i); 595 ns/op
❌stringBuffer.append(i); 2272 ns/op 3.8x slower

WHAT? 😱 How? I have no explanation at this time.

In general, explanation is: "s+= i;" is slow because the String instance is immutable in Java. Therefore it always creates a new String and copies s1 and s2. StringBuilder/Buffer are mutable. However, SBuffer has all methods synchronized, making it slower than Builder. 🤓

🤯 However, I have no idea how to explain outlier produced by 100. It's reproducible even in non parametrized benchmark and definitely a subject for future investigation and tweets.

Source as always can be found in: github.com/xp-vit/java-lo…

That's a wrap!

If you enjoyed this thread:

1. Follow me @xpvit for more Java, Cloud, and Linux knowledge.
2. RT the tweet below to share this thread with your audience

https://twitter.com/421103460/status/1621768828731494401